ml-investment 0.0.3

Machine learning tools for investment

Ml_investment

Machine learning tools for investment tasks. The purpose of these tools is to obtain deeper analytics about companies traded on the stock exchange.

Installation

PyPI package

pip install ml-investment

Source latest version

pip install git+https://github.com/fartuk/ml_investment

or

git clone https://github.com/fartuk/ml_investment
cd ml_investment
pip install .

Run tests

cd /path/to/ml_investmant
pytest

Run tests in Docker

docker build . -t tests
docker run tests

Pipelines

All investment tools represented as pipelines composed of feature and target calculation, model training and validation. Simple example of pipeline creation:

data_loader = SF1Data(config['sf1_data_path'])

fc1 = QuarterlyFeatures(
    columns=["revenue", "netinc", "debt"],
    quarter_counts=[2, 4, 10],
    max_back_quarter=5)

fc2 = BaseCompanyFeatures(
    cat_columns=["sector", "sicindustry"])

feature = FeatureMerger(fc1, fc2, on='ticker')
target = QuarterlyTarget(col='marketcap', quarter_shift=0)

model = GroupedOOFModel(LogExpModel(lgbm.sklearn.LGBMRegressor()),
                        group_column='ticker', fold_cnt=5)

pipeline = BasePipeline(feature=feature, 
                        target=target, 
                        model=model, 
                        metric=median_absolute_relative_error)

pipeline.fit(data_loader, ['AAPL', 'TSLA', 'NVDA', 'K'])
pipeline.export_core('models_data/marketcap')

Example of loading and executing pipeline

pipeline = BasePipeline.load('models_data/marketcap.pickle')
pipeline.execute(data_loader, ['INTC'])

ticker	date	y
INTC	2021-01-22	4.363793e+11
INTC	2020-10-23	2.924576e+11
INTC	2020-07-24	3.738603e+11
INTC	2020-04-24	3.766202e+11
INTC	2020-01-24	4.175332e+11

Marketcap evaluation

Model is used to estimate current fair company marketcap. Pipeline consist of calculating quarterly-based statistics of fundamental company indicators(revenue, netinc etc) and training to predict real market capitalizations. Since some companies are overvalued and some are undervalued, the model makes an average "fair" prediction.

To fit default pre-defined marketcap prediction pipeline run scripts/fair_marketcap.py:

python3 scripts/fair_marketcap.py --config_path config.json

Lower predicted marketcap may indicates that company is overvalued according its fundamentdal base.

Quarter marketcap difference

Model is used to evaluate quarter-to-quarter(q2q) company fundamental progress. Pipeline consist of calculating q2q results progress(e.g. 30% revenue increase, decrease in debt by 15% etc) and prediction real q2q marketcap difference. So model prediction may be interpreted as "fair" marketcap change according this fundamental change.

To fit default pre-defined marketcap prediction pipeline run scripts/fair_marketcap_diff.py:

python3 scripts/fair_marketcap_diff.py --config_path config.json

Similarly, a higher predicted capitalization may indicate that the company has fundamentally grown more than its value.

Future down-std prediction

Model is used to predict future down-std value. Pipeline consist of time-series model training and validation on real marketcap values. Model prediction may be interpreted as "risk" for the next quarter.

To fit default pre-defined down-std prediction pipeline run scripts/marketcap_down_std.py:

python3 scripts/marketcap_down_std.py --config_path config.json

Higher predicted down-std may be associated with higher risk.

Data

Most of feature calculators expect data_loader to have folowing structure:

class DataLoader:
    def load_base_data(self) -> pd.DataFrame:
        # returned pd.DataFrame should have ["ticker"] column
    def load_quartely_data(self, tickers: List[str]) -> pd.DataFrame:
        # returned pd.DataFrame should have ["ticker", "date"] columns
    def load_daily_data(self, tickers: List[str]) -> pd.DataFrame:
        # returned pd.DataFrame should have ["ticker", "date"] columns

There are pre-defined SF1Data class implements this structure. It is based on the data from https://www.quandl.com/databases/SF1

sf1
├── core_fundamental 
│   ├── AAPL.json
│   ├── FB.json
│   └── ...
├── daily
│   ├── AAPL.json
│   ├── FB.json
│   └── ...
└── tickers.csv